How to use this resource

The book follows the four phases of the AI lifecycle: design, development, evaluation and deployment. You can start at whichever phase your project is in.

If you are starting a new AI project¶

Start with Design. The decisions made there are the hardest to change later: who will use the system, which problem it solves, which data you need and which ethical and legal reviews apply.

If you are building the model¶

Go to Development for data management, privacy, fairness in training data and the design of how clinicians work with the tool. The FUTURE-AI principles section groups all 30 FUTURE-AI recommendations by principle instead of by phase.

If you are preparing for clinical validation¶

Evaluation covers how to plan an evaluation, validate at other sites, check fairness across patient groups and report results with guidelines such as TRIPOD+AI Collins et al. (2024) and CLAIM Mongan et al. (2020).

If you are deploying or running an AI system¶

Deployment covers integration into clinical work, quality control, monitoring, logging, regulatory compliance and governance over time.

If you are looking for a practical tool¶

The Toolbox collects workshop methods, games, impact assessments, reporting guidelines and software libraries, sorted by lifecycle phase and FUTURE-AI principle. In the chapters, a Tools for this step box points to the tools for that section.

If you are new to responsible AI¶

The Pathways section suggests a route through the book for your role.

Conventions used in this book¶

Many chapters also follow a fictional hospital that builds a sepsis early warning model. The example appears in boxes headed Running case: sepsis early warning.